The Recoil Limit and Sub-Recoil Cooling

Laser cooling lives on a single quantum fact that quantum optics hands it: a photon carries momentum $\hbar k$ , and every time an atom emits one spontaneously it recoils by exactly that much, in a random direction it cannot choose. Doppler cooling and the sub-Doppler Sisyphus mechanism both harness the average of these kicks to build a friction that drags atoms toward $v=0$ . But the same scattering that provides friction also provides irreducible heating: the last photon an atom throws off, when its velocity is already tiny, blurs its momentum by a full recoil. That single leftover kick is the floor. It defines the recoil energy $E_r=\hbar^2k^2/2m$ and the recoil temperature $T_r$ , and it says that no cooling scheme which keeps scattering light can hold a momentum spread much narrower than $\hbar k$ . For alkali atoms on their D lines this floor sits between a tenth of a microkelvin and a few microkelvin — below the temperatures sub-Doppler cooling actually reaches, so it is a real barrier in the laboratory, not a textbook abstraction. Crossing it demands inverting the whole logic of the machine. Instead of cooling atoms by making them scatter, you must make the cold atoms stop scattering: render them dark to the light precisely when their velocity approaches zero, so that once an atom random-walks into the neighbourhood of $v=0$ it can no longer be kicked, and it stays there. Two landmark techniques do exactly this — velocity-selective coherent population trapping, which parks atoms in a quantum dark superposition, and Raman cooling, which shapes a two-photon excitation spectrum with a hole at $v=0$ . Both punch the momentum width below one recoil, and both connect, surprisingly, to the mathematics of Lévy flights.

Key takeaways

Every spontaneous emission kicks an atom by one photon recoil $\hbar k$ in a random direction; the associated recoil energy $E_r=\hbar^2k^2/2m$ and recoil temperature $T_r$ set a natural floor for any cooling that relies on scattering light.
The recoil limit is deeper and more fundamental than the Doppler limit: for Rb-87 the Doppler floor is $\approx146\ \mu\mathrm{K}$ while $T_r\approx0.36\ \mu\mathrm{K}$ , several hundred times colder.
Sub-Doppler (Sisyphus) cooling reaches a few times $E_r$ but still scatters continuously, so it too saturates near the recoil scale — it cannot cross it.
To go below $T_r$ , an atom must be driven toward a low-scattering or dark state near $v=0$ , so that both the cooling force and momentum diffusion are strongly suppressed where the cooling should stop.
VSCPT parks atoms in a coherent non-coupled superposition that appears as a double-peaked momentum distribution at $\pm\hbar k$ with width $\ll\hbar k$ ; Raman cooling funnels atoms into a velocity hole at $v=0$ with tailored two-photon pulses.
Sub-recoil cooling has no steady state: the escape rate vanishes as a power of velocity, trapping times follow a Lévy law with no finite mean, and the width narrows indefinitely as $\Delta p\sim\hbar k(\tau_0/\theta)^{1/2}$ .
Key idea: a single spontaneously emitted photon imparts a recoil $\hbar k$ ; any scheme that continuously scatters light hits a floor near $T_r=\hbar^2k^2/(mk_B)$ , and sub-recoil cooling beats it only by making cold atoms dark near $v=0$ so scattering shuts off exactly where cooling should stop.

The core insight

A single spontaneously emitted photon imparts a recoil $\hbar k$ ; any cooling scheme that continuously scatters light therefore hits a floor near $T_r=\hbar^2k^2/(mk_B)$ . Sub-recoil cooling beats this only by making cold atoms dark — decoupling them from the light near $v=0$ so photon scattering shuts off exactly where cooling should stop.

Where this lives today

The sub-recoil toolkit reshaped precision measurement more than it reshaped the race to quantum degeneracy. Velocity-selective Raman pulses — the same stimulated two-photon transitions Kasevich and Chu used to cool below $T_r$ — became the beam-splitters and mirrors of light-pulse atom interferometers. In an atomic-fountain gravimeter or gyroscope, a narrow, sub-recoil velocity spread is what keeps the interferometer's arms overlapping and its fringes sharp, and the same Raman selectivity is used to prepare a single, ultra-narrow velocity class before the interferometer even begins. These devices now measure $g$ to parts in $10^9$ , sense rotation for inertial navigation, and test the universality of free fall. The recoil scale also sets the agenda for the coldest samples ever made. Free-space sub-recoil methods proved that cooling below $T_r$ was possible, but for reaching Bose-Einstein condensation the practical last stage became evaporative cooling in magnetic and optical traps, where the light is switched off and no recoil heating survives. The recoil energy returns as the thing to engineer away: in optical-lattice clocks the atoms are confined more tightly than a wavelength (the Lamb-Dicke regime), so the recoil is absorbed by the lattice and the clock transition becomes recoil-free. And on orbit, NASA's Cold Atom Laboratory on the International Space Station uses delta-kick collimation — a pulsed magnetic lens that slams a released condensate's expansion velocity toward zero — to reach effective temperatures in the picokelvin range, far below any single-photon recoil.

The physical picture

Think of laser cooling as a random walk in velocity space with a characteristic recoil scale $v_r=\hbar k/m$ . Friction supplies a gentle drift toward $v=0$ , while photon absorption and spontaneous emission produce momentum kicks whose characteristic magnitude is set by $\hbar k$ , with the spontaneous-emission direction determined randomly. While the atom is fast, the drift wins on average and the velocity spread shrinks. But as the distribution narrows toward a width of order $v_r$ , ordinary photon-scattering cooling reaches a regime where recoil heating competes with cooling. This defines the recoil scale $v_r$ and $T_r$ , but it is not an absolute floor: sub-recoil schemes can cool below this scale by suppressing scattering near selected low-velocity states.

Figure

The recoil scale in velocity space: friction drives atoms toward v = 0 while photon scattering produces recoil kicks of characteristic size vᵣ = ħk/m ≈ 5.9 mm/s for Rb-87, giving the recoil temperature scale Tᵣ ≈ 0.36 μK for ordinary scattering-based cooling

To beat the floor you must break the one assumption the argument rests on: that scattering never switches off. Imagine painting the velocity axis so that an atom sitting at $v=0$ is transparent — it simply does not absorb — while atoms with $v\neq0$ still scatter freely. Now the random walk becomes a trap. A cold atom that wanders into the transparent window stops absorbing, stops diffusing, and freezes there; a hot atom keeps scattering, keeps random-walking, until it too stumbles into the window and is caught. The narrower you cut the transparent notch, the colder the atoms you collect, because only those with velocity inside the notch survive without being reheated. This is precisely the trick that VSCPT and Raman cooling engineer, by two very different routes.

Figure

Carving a dark window: the scattering rate Γ(v) is suppressed to zero at v = 0 (green notch) so cold atoms stop diffusing and pile up, producing a sub-recoil momentum distribution — two sharp peaks at p = ±ħk of width Δp ≪ ħk (blue) instead of the recoil-limited Gaussian of width ħk (grey dashed)

VSCPT's transparency is a genuine quantum dark state. Two counterpropagating beams of opposite circular polarization couple two ground sublevels to a common excited state; a particular coherent superposition of those sublevels has two absorption amplitudes that interfere destructively, so it never absorbs. But the interference is exact only when the two momentum components are degenerate — that is, only at $p=0$ — so the dark state is velocity-selective. Raman cooling builds the same transparency dynamically instead of from interference: it drives velocity-selective two-photon transitions whose excitation spectrum is deliberately shaped to have a null at $v=0$ , then optically pumps atoms toward that null. Either way there is no steady state. The rate at which an atom leaks out of the dark region shrinks the closer it is to $v=0$ , so residence times have no finite average — they obey a broad Lévy distribution — and the momentum peak keeps sharpening the longer you wait, roughly as $1/\sqrt{\theta}$ , rather than settling to a fixed temperature.

The core equations

Everything starts from the momentum of one photon. A quantum of the cooling light carries $\hbar k$ , so absorbing or emitting it shifts the atom's momentum by that amount and its kinetic energy by the recoil energy. The three recoil scales below — energy, velocity, and frequency — are the same physical quantity expressed in the units each part of the problem naturally uses.

E_r = \dfrac{\hbar^2 k^2}{2m}, \qquad v_r = \dfrac{\hbar k}{m}, \qquad \omega_r = \dfrac{\hbar k^2}{2m} \qquad (1)

where: $E_r$ — single-photon recoil energy; $k=2\pi/\lambda$ — photon wavevector magnitude; $m$ — atomic mass; $v_r=\hbar k/m$ — recoil velocity, the speed an atom at rest gains from absorbing one photon; $\omega_r=E_r/\hbar$ — recoil frequency. Reading: $E_r$ is the kinetic energy of an atom moving at $v_r$ , and it is tiny — for Rb-87 on the D2 line, $\omega_r/2\pi\approx3.8$ kHz, far below the megahertz natural linewidth, which is exactly why the recoil floor lies below the Doppler floor. Converting the recoil energy to a temperature gives the barrier every scattering scheme runs into. Conventions differ by a numerical factor — some authors set $k_BT_r=E_r$ , others $2E_r$ — so the number must always be quoted with its definition. Here we use the common Foot convention.

k_B T_r = \dfrac{\hbar^2 k^2}{m} = 2E_r \qquad (2)

where: $T_r$ — recoil temperature; $k_B$ — Boltzmann constant. Reading: $T_r$ is the temperature at which the thermal momentum spread equals one photon recoil, $\Delta p\approx\hbar k$ . For Rb-87 this is $\approx0.36\ \mu\mathrm{K}$ (with $E_r/k_B\approx0.18\ \mu\mathrm{K}$ ); for sodium $\approx2.4\ \mu\mathrm{K}$ ; for metastable helium $\approx4.1\ \mu\mathrm{K}$ — all sub-microkelvin to microkelvin, and all below the Doppler limit. Why is $T_r$ a floor and not just another scale? Because the light that cools also heats. The friction that damps velocity and the diffusion that blurs it both come from photon scattering, and their ratio fixes the steady-state temperature — a fluctuation-dissipation balance. Since the diffusion carries the recoil scale, the balance bottoms out there.

D_p \approx \eta\,\hbar^2 k^2 \, \Gamma_{\mathrm{sc}}, \qquad k_B T_{\mathrm{ss}} = \dfrac{D_p}{m\alpha} \qquad (3)

where: $D_p$ — momentum-space diffusion coefficient; $\Gamma_{\mathrm{sc}}$ — photon scattering rate; $\eta$ — order-unity factor counting the absorption and emission recoils per cycle; $\alpha$ — velocity-damping (friction) coefficient; $T_{\mathrm{ss}}$ — steady-state temperature. Reading: momentum diffusion grows with the photon-scattering rate $\Gamma_{\rm sc}$ and carries the recoil scale $\hbar^2k^2$ . In ordinary scattering-based cooling, this recoil heating competes with friction and sets a characteristic temperature scale; sub-recoil schemes evade this balance by suppressing $\Gamma_{\rm sc}$ near the selected low-momentum dark state. To break the floor you must make the coldest atoms transparent. VSCPT does this with a coherent dark superposition of the two ground sublevels — a non-coupled state whose two absorption pathways cancel. It is written most cleanly in terms of the atom's momentum $p$ along the beam axis.

|\psi_{\mathrm{NC}}\rangle = \dfrac{1}{\sqrt{2}}\left(|g_-,\, p-\hbar k\rangle - |g_+,\, p+\hbar k\rangle\right) \qquad (4)

where: $|g_\pm\rangle$ — the two ground Zeeman sublevels driven by the $\sigma^\pm$ beams; $p$ — atomic momentum along the axis; $|\psi_{\mathrm{NC}}\rangle$ — the non-coupled superposition whose net dipole coupling to the excited state vanishes. Reading: the two components carry momenta $p\mp\hbar k$ ; their absorption amplitudes to the common excited state are equal and opposite, so the state is dark — but only at $p=0$ do the two components share a kinetic energy, and away from $p=0$ that mismatch leaks the atom back into the light. That leakage is what makes the trap velocity-selective, and its shape is what makes the cooling never stop. The escape rate vanishes with velocity, so the coldest atoms are trapped longest, and the momentum width keeps shrinking with interaction time instead of saturating.

\gamma(p)\propto p^{2}\ \ (p\to 0), \qquad \Delta p(\theta) \sim \hbar k \left(\dfrac{\tau_0}{\theta}\right)^{1/2} \qquad (5)

where: $\gamma(p)$ — rate at which an atom of momentum $p$ leaves the dark region; $\theta$ — total cooling interaction time; $\tau_0$ — characteristic single-scattering time; $\Delta p$ — width of the momentum peak. Reading: because $\gamma\to0$ as $p\to0$ , trapping times have a divergent mean (Lévy statistics) and there is no steady state; the width falls as $\theta^{-1/2}$ , so $T_{\mathrm{eff}}=(\Delta p)^2/mk_B$ is pushed further below $T_r$ the longer you cool. Raman cooling reaches the same dark-at- $v=0$ condition without a coherent dark state, using two-photon stimulated transitions between hyperfine ground levels. Their resonance condition is Doppler-sensitive, which is precisely what makes them velocity-selective.

\omega_2 - \omega_1 = \omega_0 + \mathbf{k}_{\mathrm{eff}}\!\cdot\!\mathbf{v} + \dfrac{\hbar k_{\mathrm{eff}}^2}{2m} \qquad (6)

where: $\omega_{1,2}$ — frequencies of the two Raman beams; $\omega_0$ — ground-state hyperfine splitting; $\mathbf{k}_{\mathrm{eff}}=\mathbf{k}_1-\mathbf{k}_2$ — effective (difference) wavevector; $\mathbf{v}$ — atomic velocity. Reading: the Doppler term $\mathbf{k}_{\mathrm{eff}}\cdot\mathbf{v}$ ties the two-photon resonance to velocity, so a pulse of finite bandwidth addresses a chosen velocity class; shaping a sequence of such pulses to leave a hole at $v=0$ funnels atoms toward zero velocity and leaves them dark once they arrive.

Derivation AThe recoil limit from momentum diffusion

The goal is to show, quantitatively, why any cooling scheme that keeps scattering light saturates at the recoil scale, no matter how clever its friction. We model laser cooling as Brownian motion in velocity space: a deterministic friction force drags the atom toward zero velocity, while the random arrival and departure of photons acts as a fluctuating force that heats. Balancing the two in steady state — the fluctuation-dissipation theorem in its cooling guise — yields a temperature equal to diffusion over friction, and the recoil energy sits inside the diffusion in a way that cannot be scaled away while photons are still being scattered.

Step 1. Near resonance the two cooling beams produce a velocity-dependent restoring force $F=-m\alpha v$ , where $\alpha$ is the damping rate set by the detuning and intensity. Why: this is the friction that makes molasses cool at all — it is the average, over many scattering events, of the momentum removed per second.

Step 2. The same scattering that averages to friction also fluctuates. Each scattered photon delivers a random $\hbar k$ from absorption and another random $\hbar k$ from spontaneous emission, so the momentum executes a random walk with diffusion coefficient $D_p=\eta\,\hbar^2k^2\,\Gamma_{\mathrm{sc}}$ , with $\eta$ an order-unity geometric factor. Why: heating is not a defect of imperfect optics — it is the shot noise of the discrete, randomly directed photons that produce the friction in the first place.

Step 3. In steady state the energy pumped in by diffusion equals the energy removed by friction, $\tfrac12 m\langle v^2\rangle_{\mathrm{ss}}=\tfrac12 k_BT_{\mathrm{ss}}$ with $k_BT_{\mathrm{ss}}=D_p/(m\alpha)$ . Why: setting the Fokker-Planck heating and cooling rates equal is exactly fluctuation-dissipation — temperature is diffusion divided by friction.

Step 4. Both $\alpha$ and $D_p$ derive from the same scattering rate — $\alpha$ from the slope of $\Gamma_{\mathrm{sc}}(v)$ , $D_p$ from its value — so their ratio cannot be pushed to zero. When the spread narrows to about one recoil, $\langle p^2\rangle\gtrsim(\hbar k)^2$ , the last, unavoidable photon dominates the balance and $k_BT_{\mathrm{ss}}\gtrsim\hbar^2k^2/m=k_BT_r$ . Why: this is the recoil floor — the cooling and the heating are the same physical process, so you cannot have one without the other down to the single-photon scale.

Step 5. The only escape route is to violate the premise: make $\Gamma_{\mathrm{sc}}(v\to0)\to0$ , so that $D_p$ vanishes locally at $v=0$ and cold atoms simply stop being kicked. Why: this is the design principle behind every sub-recoil method — kill the diffusion where you want the atoms to accumulate. *The Rb-87 numbers confirm the hierarchy: $T_D\approx146\ \mu\mathrm{K}$ and $T_r\approx0.36\ \mu\mathrm{K}$ , so $T_r\ll T_D$ , consistent with scattering-based cooling saturating at the recoil scale.*

Go deeper · Derivation B · 2 minVelocity selectivity of the VSCPT dark state

Derivation BVelocity selectivity of the VSCPT dark state

The goal is to prove that the coherent superposition VSCPT relies on is decoupled from the light exactly at $p=0$ and only there, so that it functions as a momentum-selective trap whose captured atoms sit at $p=\pm\hbar k$ with a width that shrinks over time. We take a $J=1\to J=1$ (or $J=1\to J=0$ ) transition driven by two counterpropagating $\sigma^+/\sigma^-$ beams, build the non-coupled state by hand so its absorption amplitudes cancel, and then show that the atomic kinetic energy provides a residual coupling back to the bright state that switches off only at zero momentum.

Step 1. Form the superposition $|\psi_{\mathrm{NC}}\rangle=\tfrac{1}{\sqrt2}(|g_-,p-\hbar k\rangle-|g_+,p+\hbar k\rangle)$ , in which the momentum labels track the photon each sublevel would have absorbed. Why: the momentum bookkeeping matters — a ground sublevel that absorbs a $\sigma^+$ photon from one beam lands in the excited state with a specific momentum, and we are pairing the two paths that reach the same excited state.

Step 2. Compute the coupling to the common excited state $|e,p\rangle$ : the two dipole amplitudes are equal in magnitude and opposite in sign, so $\langle e,p|\hat V|\psi_{\mathrm{NC}}\rangle\propto d_--d_+=0$ . Why: destructive interference of the two absorption pathways means no light is absorbed — and no absorption means no spontaneous emission, hence no recoil heating at all.

Step 3. Now let the atom move. The two components carry momenta $p\mp\hbar k$ , so the kinetic Hamiltonian $\hat p^2/2m$ gives them different energies unless $p=0$ . This energy difference, of order $2\hbar k p/m$ , couples $|\psi_{\mathrm{NC}}\rangle$ to the orthogonal bright state $|\psi_{\mathrm{C}}\rangle$ . Why: darkness is not automatic once the atom moves — motion mixes the dark and bright states, and only at $p=0$ does the mixing vanish.

Step 4. The leak rate out of the dark state therefore vanishes quadratically near zero momentum, $\gamma(p)\sim\gamma_0(p/\hbar k)^2$ for small $p$ . Why: this quadratic suppression makes the dark-state region strongly velocity-selective — the closer an atom is to $p=0$ , the longer it remains dark.

Step 5. Atoms accumulate in the non-scattering region around $p=0$ and, because the two components live at $p\mp\hbar k$ , they appear as twin peaks at $p=\pm\hbar k$ in the state-resolved distribution; longer interaction times narrow those peaks below one recoil. Why: this is the experimental signature — a double-peaked momentum profile whose peaks are narrower than $\hbar k$ , impossible for any scattering-limited cooling. *The 1988 experiment by Aspect et al. cooled metastable helium to $\sim T_r/2$ , producing a two-peaked momentum distribution of width well below $\hbar k$ .*

Go deeper · Derivation C · 2 minLevy statistics and unbounded narrowing

Derivation CLevy statistics and unbounded narrowing

The goal is to explain the strangest feature of sub-recoil cooling: it never reaches a steady state, and its momentum width keeps decreasing without bound as the interaction time grows. We treat the cooled atom as a walker that alternates between a bright phase, where it random-walks in velocity, and a trapping phase near $v=0$ , where the escape rate vanishes as a power of velocity. Because the deepest-trapped atoms escape most slowly, the distribution of trapping times acquires a heavy power-law tail whose mean diverges, and standard statistical equilibrium — which assumes a finite mean waiting time — simply does not apply.

Step 1. Near zero velocity the escape rate behaves as $\gamma(v)\simeq\gamma_0(v/v_r)^2$ , the quadratic law established for both VSCPT and Raman cooling. Why: this power law is the whole engine — it means an atom that wanders very close to $v=0$ can stay trapped for an enormous time.

Step 2. Inverting the escape rate, the trapping times are distributed as $P(\tau)\sim\tau^{-1-\mu}$ with $0<\mu<1$ , a broad Lévy tail. Because $\mu<1$ the mean trapping time $\langle\tau\rangle$ diverges. Why: a divergent mean means there is no characteristic time in the problem — the statistics are dominated by rare, extremely long trapping events, not by a typical one.

Step 3. In the idealized model, the absence of a finite mean waiting time prevents relaxation to an ordinary stationary distribution. Instead, as cooling continues, an increasing fraction of atoms occupies an increasingly narrow velocity class around $v=0$ . Why: this anomalous, time-dependent narrowing is a characteristic prediction of the idealized sub-recoil model rather than a universal property of every experimental cooling scheme.

Step 4. Carrying the idealized Lévy analysis through gives the width $\Delta p(\theta)\sim\hbar k(\tau_0/\theta)^{1/2}$ for the assumed $\gamma(p)\propto p^2$ escape-rate law, so the momentum peak narrows algebraically with interaction time rather than approaching a fixed recoil-limited width. Why: this $\theta^{-1/2}$ narrowing is the operational fingerprint that distinguishes sub-recoil cooling from ordinary thermalization, which would freeze at a fixed temperature. *This Lévy-flight picture was developed by Bardou, Bouchaud, Aspect and Cohen-Tannoudji (1994) and elaborated in their monograph Lévy Statistics and Laser Cooling.*

Limiting cases

Where the successive temperature scales sit for an alkali atom being laser cooled, from the coarsest floor to the finest:

Case	Result	Meaning
Doppler limit	$k_B T_D = \dfrac{\hbar\Gamma}{2}$	Set by the natural linewidth $\Gamma$ ; hundreds of $\mu$ K for alkalis ( $\approx146\ \mu$ K for Rb-87). The floor for simple two-level molasses.
Sub-Doppler (Sisyphus)	$k_B T \sim \dfrac{\hbar\Omega^2}{\lvert \delta\rvert }\;\gtrsim\; \text{few}\times E_r$	Multilevel light shifts reach a small multiple of the recoil energy, but continuous scattering keeps it above $T_r$ .
Recoil limit	$k_B T_r = \dfrac{\hbar^2 k^2}{m}$	Set by single-photon recoil; sub- $\mu$ K for alkalis ( $\approx0.36\ \mu$ K Rb-87). Below $T_D$ whenever $\Gamma\gg\hbar k^2/m$ , which holds for allowed dipole lines.
Sub-recoil (VSCPT / Raman)	$k_B T_{\mathrm{eff}} \ll k_B T_r$	Momentum width $\Delta p\ll\hbar k$ ; narrows as $\theta^{-1/2}$ without bound because there is no true steady state.
Tight confinement (Lamb-Dicke)	recoil absorbed by trap	In a lattice tighter than $\lambda/2\pi$ the recoil is taken up by the trap, not the atom; the carrier line becomes recoil-free (optical-clock regime).

Common trapSub-recoil cooling reaches an ever-lower steady-state temperature

Two traps hide in that sentence. First, there is no steady state: the escape rate from the dark region vanishes with velocity, so the trapping-time distribution has a divergent mean (Lévy statistics) and the momentum width keeps shrinking as $\theta^{-1/2}$ for as long as you cool. Quoting a single "temperature" is a convenient summary of a non-Gaussian, non-equilibrium distribution — useful, but not a true thermodynamic temperature. Second, the recoil temperature itself is convention-dependent: different texts set $k_BT_r$ equal to $E_r$ , to $2E_r$ , or to $\hbar^2k^2/m$ . The physics is identical, but numerical statements differ by factors of two, so always state which convention you are using before comparing $T_r$ to a measured width. And do not confuse the recoil limit with the Doppler limit — they are different scales set by different physics ( $\hbar k^2/m$ versus $\Gamma$ ), and for allowed dipole lines the recoil floor lies far below the Doppler floor.

Worked examples

1 · The recoil scales and the floor for rubidium-87 (confidence builder)

Rubidium-87 is cooled on its D2 line, $\lambda=780.24$ nm, with natural linewidth $\Gamma=2\pi\times6.07$ MHz and mass $m=87\times1.66\times10^{-27}$ kg $=1.44\times10^{-25}$ kg. Find the recoil velocity, recoil energy, recoil frequency, and recoil temperature, and compare the recoil temperature with the Doppler limit.

Step 1. The wavevector is $k=2\pi/\lambda=2\pi/(780.24\times10^{-9})=8.05\times10^{6}\ \mathrm{m}^{-1}$ , so one photon momentum is $\hbar k=1.055\times10^{-34}\times8.05\times10^6=8.49\times10^{-28}\ \mathrm{kg\,m/s}$ .

Step 2. The recoil velocity is $v_r=\hbar k/m=8.49\times10^{-28}/1.44\times10^{-25}=\mathbf{5.9\times10^{-3}\ m/s}$ , about 5.9 mm/s — the speed a rubidium atom at rest picks up from a single photon.

Step 3. The recoil energy is $E_r=\hbar^2k^2/2m=(8.49\times10^{-28})^2/(2\times1.44\times10^{-25})=2.50\times10^{-30}\ \mathrm{J}$ . As a frequency, $\omega_r/2\pi=E_r/h=\mathbf{3.8\ kHz}$ ; as a temperature, $E_r/k_B=2.50\times10^{-30}/1.38\times10^{-23}=\mathbf{0.18\ \mu K}$ .

Step 4. In the Foot convention $k_BT_r=2E_r$ , so $T_r=2\times0.18=\mathbf{0.36\ \mu K}$ . The Doppler limit is $k_BT_D=\hbar\Gamma/2$ , giving $T_D=\hbar\times(2\pi\times6.07\times10^6)/(2k_B)=\mathbf{146\ \mu K}$ .

Step 5. The ratio is $T_D/T_r=146/0.36\approx\mathbf{400}$ . The recoil floor sits several hundred times below the Doppler floor — decisively deeper — which is why sub-Doppler cooling can beat $T_D$ yet still stall near $T_r$ , and why breaking $T_r$ needs an entirely different mechanism. v_r ≈ 5.9 mm/s, E_r/k_B ≈ 0.18 μK, ω_r/2π ≈ 3.8 kHz, T_r ≈ 0.36 μK, about 400× below T_D ≈ 146 μK.

2 · How far below the recoil limit does VSCPT reach, and how long does it take?

Metastable helium ( $^4\mathrm{He}^\ast$ , $\lambda=1083$ nm, $m=4$ amu) is cooled by one-dimensional VSCPT. Its recoil temperature is $T_r\approx4.1\ \mu$ K and recoil velocity $v_r\approx9.2$ cm/s. The 1988 experiment reached a momentum width $\Delta p\approx\hbar k/2$ . (a) What effective temperature does that width correspond to, in units of $T_r$ ? (b) By what factor must the interaction time $\theta$ grow to halve the width again?

Step 1. The effective temperature from a momentum width is $k_BT_{\mathrm{eff}}=(\Delta p)^2/m$ , so $T_{\mathrm{eff}}/T_r=(\Delta p/\hbar k)^2$ using $k_BT_r=(\hbar k)^2/m$ . This is the clean way to read a width as a temperature.

Step 2. With $\Delta p=\hbar k/2$ , $T_{\mathrm{eff}}/T_r=(1/2)^2=1/4$ , so $T_{\mathrm{eff}}\approx0.25\times4.1=\mathbf{1.0\ \mu K}$ — already a factor of four below the recoil temperature, impossible for any scattering-limited scheme.

Step 3. Sub-recoil narrowing follows $\Delta p\propto\theta^{-1/2}$ . To halve $\Delta p$ requires $(\theta_2/\theta_1)^{-1/2}=1/2$ , i.e. $\theta_2/\theta_1=\mathbf{4}$ : four times the interaction time buys a factor-two-narrower peak and a factor-four-lower $T_{\mathrm{eff}}$ .

Step 4. This diminishing return — cooling that pays only as $\theta^{-1/2}$ — is the practical signature of sub-recoil methods and the reason there is no fixed endpoint: the temperature is set by how long you are willing to wait, not by an equilibrium. T_eff ≈ T_r/4 ≈ 1.0 μK at Δp = ħk/2, and halving the width again costs 4× the interaction time.

Physics of the portrait

Sub-recoil cooling has one presiding figure. It was in Claude Cohen-Tannoudji's group at the École Normale Supérieure that the first technique to cross the recoil limit was invented, understood, and then reframed in the unexpected language of Lévy flights — a rare case of an experiment sending physicists back to rethink a piece of pure probability theory.

Origins · 1 minClaude Cohen-Tannoudji and how this was discovered

Claude Cohen-Tannoudji

1933–2025 · laser cooling and coherent atom-light interaction

Claude Cohen-Tannoudji spent his career building the theoretical machinery that turns the messy interaction of atoms and light into something calculable — the dressed-atom picture, the optical Bloch equations, and the systematic accounting of light shifts and radiative forces that underpins all of laser cooling. In 1988, with Alain Aspect, Ennio Arimondo, Robert Kaiser and Nathalie Vansteenkiste, his group demonstrated velocity-selective coherent population trapping on metastable helium, the first cooling of neutral atoms below the single-photon recoil limit — parking atoms in a coherent dark superposition that could not absorb, and reading it out as the now-iconic double-peaked momentum distribution at $\pm\hbar k$ . What might have been a laboratory curiosity became something deeper when, in 1994, Bardou, Bouchaud, Aspect and Cohen-Tannoudji recognized that the divergent trapping times of the coldest atoms are governed by Lévy statistics, so that sub-recoil cooling has no steady state and narrows without bound — a genuine two-way traffic between quantum optics and the theory of anomalous diffusion. He shared the 1997 Nobel Prize in Physics with Steven Chu and William Phillips for the development of methods to cool and trap atoms with laser light, the toolkit that opened the door to Bose-Einstein condensation, atom interferometry, and optical-lattice clocks.

History

1985

— Chu and coworkers at Bell Labs demonstrate optical molasses, Doppler-cooling sodium in three dimensions

1988

— Aspect, Arimondo, Kaiser, Vansteenkiste and Cohen-Tannoudji cool metastable helium below the recoil limit by VSCPT

1992

— Kasevich and Chu demonstrate velocity-selective Raman cooling of sodium below the recoil temperature

1994

— Bardou, Bouchaud, Aspect and Cohen-Tannoudji recast sub-recoil cooling as a Lévy-flight problem with no steady state

1997

— Chu, Cohen-Tannoudji and Phillips share the Nobel Prize for laser cooling and trapping of atoms

Going further

Texts. Foot, Atomic Physics (2005), Ch. 9 (laser cooling, the recoil and sub-recoil limits) · Metcalf and van der Straten, Laser Cooling and Trapping (1999), Ch. 8 (VSCPT and Raman cooling) · Cohen-Tannoudji, Dupont-Roc and Grynberg, Atom–Photon Interactions (1992) · Demtröder, Laser Spectroscopy, Vol. 2 (cooling and trapping) · Loudon, The Quantum Theory of Light, 3rd ed. (spontaneous emission and recoil). Key numerical benchmarks: $E_r=\hbar^2k^2/2m$ and $T_r=\hbar^2k^2/(mk_B)$ evaluate to $E_r/k_B\approx0.18\ \mu$ K, $T_r\approx0.36\ \mu$ K, $v_r\approx5.9$ mm/s, $\omega_r/2\pi\approx3.8$ kHz for Rb-87 at 780.24 nm; $E_r/k_B\approx1.2\ \mu$ K, $T_r\approx2.4\ \mu$ K for sodium at 589 nm; and $T_r\approx4.1\ \mu$ K for $^4\mathrm{He}^\ast$ at 1083 nm. The Doppler limit for Rb-87 is $T_D=\hbar\Gamma/2k_B\approx146\ \mu$ K ( $\Gamma/2\pi\approx6.07$ MHz). The steady-state temperature follows from the diffusion-over-friction balance. The temperature-convention caveat ( $k_BT_r=E_r$ versus $2E_r$ ) is flagged in the pitfall box above.

Prerequisites. Radiation Pressure and the Scattering Force · Doppler Cooling and the Doppler Limit · Sub-Doppler Cooling and the Sisyphus Mechanism · Optical Molasses. Leads to. Evaporative Cooling · Bose-Einstein Condensation in Dilute Gases · Magnetic Trapping of Neutral Atoms · Degenerate Fermi Gases and the BEC-BCS Crossover.

Ask the tutor: Ask me to walk through why the very last scattered photon — not the linewidth — is what sets the recoil floor, and how a dark state sidesteps it.

Is the recoil limit the same as the Doppler limit, just for a different atom?

No — they are set by completely different physics. The Doppler limit $k_BT_D=\hbar\Gamma/2$ comes from the natural linewidth $\Gamma$ and the balance of friction against scattering in a two-level atom; it is typically hundreds of microkelvin. The recoil limit $k_BT_r=\hbar^2k^2/m$ comes from the momentum of a single photon and is set by $\hbar k^2/m$ , independent of the linewidth. For any allowed dipole transition $\Gamma\gg\hbar k^2/m$ , so the recoil floor lies far below the Doppler floor — about 400 times lower for Rb-87. That gap is exactly why sub-Doppler cooling can leap below $T_D$ and still find itself stuck above $T_r$ .

If sub-Doppler cooling already beats the Doppler limit, why does it stop at the recoil scale?

Because sub-Doppler (Sisyphus) cooling changes the friction, not the fundamental heating. It uses polarization gradients and light shifts to remove energy far more efficiently than simple Doppler friction, so it can reach a few times the recoil energy. But it still works by scattering photons, and every scattered photon still delivers a random $\hbar k$ recoil. Once the velocity spread narrows to about one recoil, that unavoidable last kick reheats the atom as fast as the friction cools it, and the temperature saturates near $T_r$ . Beating $T_r$ requires turning the scattering off for cold atoms — something Sisyphus cooling, which needs continuous scattering to function, cannot do.

What actually makes an atom "dark" in VSCPT?

Quantum interference between two absorption pathways. Two counterpropagating beams of opposite circular polarization couple two ground sublevels to a common excited state. A particular coherent superposition of those sublevels — the non-coupled state — has two absorption amplitudes that are equal and opposite, so they cancel: the atom cannot absorb, and what it cannot absorb it cannot spontaneously re-emit, so it feels no recoil. The catch is that this cancellation is exact only when the two momentum components share the same kinetic energy, which happens solely at $p=0$ . Away from zero velocity the kinetic-energy mismatch mixes the dark state back into the bright state, and the atom starts scattering again. That is what makes the darkness velocity-selective.

Why does the VSCPT momentum distribution have two peaks instead of one at v = 0?

Because the dark state is not a single momentum state — it is a superposition of two, at $p-\hbar k$ and $p+\hbar k$ . When you accumulate atoms in the non-coupled state around $p=0$ and then measure momentum, resolving the internal state, the two components show up as separate peaks at $p=+\hbar k$ and $p=-\hbar k$ . Each peak is much narrower than $\hbar k$ , because only atoms within a tiny velocity window of $v=0$ stay dark long enough to survive. So the twin peaks are not two temperature classes; they are the two momentum halves of one coherent dark state, and their sub-recoil width is the real signature of having beaten the recoil limit.

How can a momentum width keep shrinking forever — doesn't cooling reach equilibrium?

Ordinary cooling reaches equilibrium because there is a typical waiting time between events, and the system relaxes to a steady distribution. Sub-recoil cooling breaks that assumption. The rate at which an atom escapes the dark region vanishes as its velocity approaches zero (as $v^2$ ), so the coldest atoms have enormously long trapping times — so long that the average trapping time diverges. With no finite mean waiting time (Lévy statistics), there is no steady state to relax into. Instead the width narrows as $\theta^{-1/2}$ : cool for four times as long and the peak is half as wide. The endpoint is set by patience and by practical limits like collisions or gravity, not by an equilibrium temperature.

If sub-recoil cooling can reach nanokelvin, why is evaporation used to make a condensate?

Because phase-space density, not just temperature, decides degeneracy — and sub-recoil methods cool without compressing. VSCPT and Raman cooling narrow the velocity spread of a dilute, free-space cloud, which is superb for making a bright, velocity-narrow atomic source, but they do not by themselves pile enough atoms into a small enough volume to cross the condensation threshold. Evaporative cooling, performed in a magnetic or optical trap with the light switched off, throws away the hottest atoms and lets the rest rethermalize, raising phase-space density by orders of magnitude with no recoil heating at all. In practice, laser cooling (often to sub-Doppler temperatures) loads the trap and evaporation finishes the job — the recoil limit set the stage, evaporation took the last step.